Computer-Aided Classification of Malignant and Benign Lesions on Mammograms
Annual rept. 1 May 98-30 Apr 99
MICHIGAN UNIV ANN ARBOR
Pagination or Media Count:
The purpose of this project is to develop computerized classification methods for mammographic abnormalities which will aid radiologists in deciding whether a patient should be biopsied. The regions of interest ROIs will be identified by radiologists, and the features to be used in classification will be computer-extracted image features. In the third year of our project, we developed a segmentation method for delineating boundaries of mammographic masses. New morphological features were extracted from these boundaries. The accuracy of segmentation and the discrimination ability of the extracted morphological features were demonstrated on a data set of 249 biopsy-proven masses. To demonstrate the generalizability of our classification method, a classifier was trained on 301 masses and was tested on 91 independent masses. The classification accuracy on the independent test set AzO.82 was close to that of an experienced breast radiologist AzO.88. Morphological features were also extracted for classification of microcalcifications. Their classification accuracy was evaluated on a data set of 145 biopsy proven microcalcifications. The combination of texture and morphological feature spaces for classification of microcalcifications as malignant or benign was also investigated.
- Medicine and Medical Research